A Model Based On Bayesian Regularization BP Neural Network To Predict The Probability Of The Listed Companies' Financial Crisis

Artificial neural network(ANN)is a rising borderline science. It is an information deal-with system invoked by biology neural network for its structure, function and some basic characters, but being abstract and simplified. ANN has distributed storage and parallel disposing form of information. Compared to the traditional statistics method, it doesn't need exact mathematical model and doesn't have any consumption (such as distribution and independence) demanding the variables to meet . It can make up the deficiency of mathematical statistics methods and gives an idea and method to solving the problems.This study focuses on the back propagation network(BP network) which is a widely applied model in the pre-warning modes in financial crisis field. The fundamental of BP network and BP algorithm was introduced and the improved algorithm was given. To counter the over-fitting problem in BP algorithm and its improvements, we proposed method which was raising the generalization of the network of BP. Based on Bayesian-regularization BP algorithm is put forward. The performance index includes a long interpreted as the long prior probability distribution over the network parameters and a s(?)m-squared error term interpreted as the log likelihood for a noise model. The former is used for penalizing the network complexity . The optimal network parameters can be obtained with the maximum posterior probabilities of the models.On the basis of study on ANN theory and realization method, applied BBP model to predict the probability of the financial crisis, at the same time other networks were built for comparation with the network we have set. The result showed that BBP has better than conventional BP arithmetic in the same net-scale and training-error. The correlative coefficient R= 0.849, and the predicting precision was 79.3% for the testing set. The models can not only exactly imitate training valuation but also make prediction accurately, the forecasting result was more precise and defective than traditional methods.our results demonstrate that Bayesian regularization BP neural network can avoid effectively over-fitting in neural network training, the algorithm products networks which have excellent generalization capabilities and can be used as a conduct in application.